KJedi: a local, open-source contract review tool for California AI-focused SaaS attorneys

KJedi is a small open-source Next.js web app, built by Marc Hoag, that runs only on your laptop (localhost). You upload a contract (PDF or DOCX), chat with Claude about the terms, and generate a red/yellow/green review memo at the end. No hosted service, no database, no logs, no analytics. Your Anthropic API key lives in .env.local and never touches the browser. Document text and chat history are held in memory only and clear when you close the tab. The system prompt is a plain TypeScript file at lib/system-prompt.ts that you can read, edit, and fork. Repo: github.com/marchoag/kjedi.

Five practical paths, each trading off privilege, cost, and effort differently: (1) consumer chat tools (ChatGPT free/Pro, Claude.ai Pro) — cheapest, but governed by Consumer Terms that include training opt-outs and weaker retention guarantees, so not privilege-safe; (2) commercial API plus your own wrapper — privilege-safe via Anthropic/OpenAI Commercial Terms (no training, DPA included), highest build effort, typically $100-500/month in API costs; (3) managed legal-AI SaaS platform — privilege-safe, lowest effort, ~$500/month and up; (4) AI-in-a-box appliance where data never leaves the device — medium effort plus hardware, ~$500-700/month equivalent; (5) on-prem frontier model on your own hardware — highest effort, $10K+ in hardware. KJedi is path #2 packaged as open source.

California AI and SaaS tech-transactions practice. The system prompt knows about SaaS subscription agreements, MSAs, SOWs, DPAs, order forms, AI-specific clauses (training data rights, output ownership, IP indemnity for AI outputs, model substitution, accuracy disclaimers, prompt/output confidentiality, upstream-model-provider disclosure), vendor agreements with AI components (OpenAI, Anthropic, AWS Bedrock, Google Vertex), privacy law (CCPA/CPRA, GDPR, DPAs), IP assignments, contractor and employee IP/confidentiality, NDAs, and open-source compliance plus AI training-data provenance.

Real estate, M&A, securities and financing, litigation, and regulated-industry compliance (healthcare, FinReg). KJedi will flag if a contract is outside its lane, but you'll get better results from a tool tuned for those areas. The system prompt at lib/system-prompt.ts is the place to fork if your practice differs.

A typical contract review session — upload, several follow-up questions, generate memo — should run just a few dollars in Anthropic API costs. Steady client volume could therefore cost a few hundred dollars per month, putting it in the same price band as managed legal-AI platforms but with the wrapper logic and system prompt fully in your control. Frontier intelligence at attorney-workload volume costs real money to deliver under any model; the legitimate options all cluster in roughly the same range.

Three things. (1) Your Anthropic account must be on Commercial Terms, not Consumer Terms. API access via the Anthropic console defaults to Commercial Terms once you add a payment method; Claude.ai Pro/Max is governed by Consumer Terms. Confirm in your console. (2) This is open-source experimental software, not a vetted vendor product. Read the code, audit the dependencies, verify the system prompt is adequate for your needs, and verify your own machine's configuration. (3) Treat KJedi's output as a junior associate's first pass, never the final word. Verify every citation before relying on it, including text inside the system prompt itself.